Relative depth prediction on Hypersim (val) benchmark leaderboard · SOTA2 Research

Relative depth prediction on Hypersim (val)

0.375δ1Blind Guess

Evaluation Results

Method	Links
Blind Guess2025.07		0.375	0.628	0.773	0.25	54.24	0.758
Claude 3.5 Sonnet2025.07		0.428	0.693	0.83	0.49	68.09	0.654
Qwen2-VL2025.07		0.432	0.698	0.831	0.44	65.88	0.638
Gemini 1.5 Pro2025.07		0.458	0.709	0.835	0.51	66.78	0.628
Llama 3.22025.07		0.458	0.711	0.835	0.53	67.51	0.608
GPT-4o2025.07		0.461	0.716	0.84	0.54	70.59	0.618
Gemini 2.0 Flash2025.07		0.461	0.715	0.839	0.59	71.11	0.615
o4-mini2025.07		0.467	0.718	0.841	0.58	74.08	0.595
4M-212025.07		0.565	0.774	0.865	0.81	88.25	0.529
Omnidata2025.07		0.568	0.772	0.864	0.81	93.74	0.528
Oracle2025.07		0.571	0.774	0.863	0.83	100	0.528
Moge2025.07		0.576	0.774	0.861	0.82	92.99	0.505
Lotus2025.07		0.578	0.779	0.866	0.82	92.59	0.529
4M-212025.07		0.636	0.814	0.888	0.89	—	0.406
Omnidata2025.07		0.768	0.867	0.911	0.95	—	0.375
Lotus2025.07		0.776	0.861	0.913	0.93	—	0.334
Moge2025.07		0.795	0.859	0.903	0.9	—	0.401